A CrewAI agent that meets a CAPTCHA needs a tool, and three CrewAI settings decide whether it is safe: the tool cache, the usage limit and the retry path. Here is a CapSkip tool built around them, with every guard attacked.

DataDome wants a cookie, not a token, and Guzzle throws on the 403 that carries the challenge. Here is the PHP that builds the captcha_url, calls CapSkip dataDome(), and keeps the cookie, User-Agent and proxy together in one Guzzle client.

DataDome answers a doubtful request with a 403 block page and wants a cookie, not a token. Here is the Node.js loop with fetch that builds the captcha_url, calls CapSkip dataDome(), and sends the cookie back with the User-Agent and address it belongs to.

AWS WAF answers with HTTP 405 and a puzzle, and solving it yields a voucher pair. Here is the C# loop with HttpClient that trades it for the cookie.

DataDome answers a doubtful request with a 403 block page and wants a cookie, not a token. Here is the C# loop that builds the captchaUrl, calls CapSkip DataDomeAsync, and sends the cookie back through HttpClient with the User-Agent and address it belongs to.

DataDome answers a doubtful request with a 403 block page and wants a cookie, not a token. Here is the Python loop that builds the captcha_url, calls CapSkip datadome(), and sends the cookie back with the User-Agent and address it belongs to.

有些页面不显示 405 页面，而是在表单里绘制 AWS WAF 拼图。下面是处理这种组件的 Node.js 循环：找到 apiKey 和 jsapi.js，调用 CapSkip 的 amazonWaf，用 voucher 换取 token，再提交表单。

AWS WAF 会用 HTTP 405 和一道拼图来应答请求，而答案是一对 voucher，不是 token。下面是一个 Python 循环：读取页面，调用 CapSkip 的 amazon_waf，用 voucher 换取 aws-waf-token cookie，然后重试。

对每个被拦截的页面都去识别，只会浪费识别次数；每遇到验证码就轮换代理，又会白白烧掉 IP。下面是一个 Python 分类器：它读取状态码、响应头和页面标记，再告诉 CapSkip 何时该识别、何时该退避。

把 v3 密钥发给 v2 方法，要等到识别结束才会发现出错。下面是一个 Python 检测器：它读取页面，判断验证码类型，提取 sitekey，并调用匹配的 CapSkip 方法，无法判断时会明确告诉你。